[EM] Unbias by definition. Quantitative bias comparisons by testing.

[Michael Ossipoff]

I've told why my methods are unbiased according to the popular meaning of 
bias. (BF's unbias depends on the distribution; Weighted BF's unbias depends 
on the accuracy of its distribution approximation). To disagree with those 
claims, tell why you disagree with that bias definition, and provide a 
substitute.

The free-seat requirement may have caused BF to look worse in the tests. The 
free seat makes the correlation result less meanngful, for judging methods' 
bias, and for judging the unproportionality of the seat allocations among 
the seat-qualified states.  That's why no-free-seat correlation results 
should always be reported too.

By the popular bias definition, bias should be empirically tested by 
correlation between the q and s/q of _cycles_, not individual states. That 
sounds radical, but it follows from the accepted bias definition. It sounds 
self-serving, coming from me, but I chose my methods because I wanted unbias 
by that definition.

I don't know how much the difference between cycle-correlation and 
state-correlation could affect how good the methods look, but I suggest that 
test reports report cycle-correlation too.

Mike Ossipoff